Flight and maintenance scheduling pose conflicting objectives: while maintenance is vital for ensuring aircraft airworthiness, it comes at the cost of taking aircraft out of operation. In current operations, airlines manually handle tail assignment and maintenance task scheduling separately, missing an opportunity to strike a better balance. This division leads to wasted maintenance resources, restricted fleet availability for schedule flexibility, inconsistent planning, and neglect of schedule resilience. This study presents a novel approach that integrates tail assignment and maintenance scheduling into a unified decision-support framework. An integer program, tailored to meet airline-specific requirements and constraints, is combined with an innovative time-space network (TSN). The TSN incorporates two distinct spaces for maintenance and network activities. The primary objective is to generate feasible plans that increase schedule efficiency (i.e., no cancellations, high fleet availability, high fleet health, and optimal use of maintenance resources) and schedule stability (i.e., limited number of late arrival disruptions during operations) the day before operation. Additionally, this framework addresses overlooked aspects in the literature: it treats maintenance tasks as variable interval activities based on aircraft-specific needs, departing from the traditional fixed interval approach. The performance of the framework is tested with real-data provided by a major European single hub-to-spoke airline, with a heterogeneous fleet of over 50 wide-body aircraft. Historical data from arrival delays is used to create robust buffers that mitigate delay propagation. A 17% reduction in maintenance time was achieved compared to the airline’s current plans, resulting in a 10% increase in fleet availability on the day of operations. This improvement is attributed to higher labour and task interval utilization, indicating the framework’s superior efficiency in scheduling maintenance tasks. Lastly, the framework produced plans more resilient to arrival delays, reducing the number of disruptions and delay propagation over 40%. This framework can be used as a decision-support tool for airlines, enabling the creation of schedules that are both robust against delays and optimized for fleet utilization.

Air traffic delays have a major impact on the aviation industry, affecting airlines, passengers, and the broader ecosystem. With increasing regulatory and sustainability pressures, accurate delay predictions are critical as they allow for precise determination of the contingency and discretionary fuel required for flights. This research aims to develop an explainable supervised learning model to improve existing en route delay predictions, focusing on intercontinental flights from North America to Amsterdam Schiphol Airport. While prior studies have explored flight delay prediction, they have not addressed two critical research gaps identified in this research: the inclusion of day-of-operations features, such as passenger information, aircraft weights, and cost index, and the use of transatlantic flight data for predictions 90 minutes before departure. To address these gaps, two Gradient-Boosted models, CatBoost and LightGBM, were trained using internal airline, airport, and METAR data. Both models outperformed the airline’s current in-use statistical model, with CatBoost achieving an MAE of 3.44 minutes and RMSE of 4.61 minutes and LightGBM achieving an MAE of 3.43 minutes and RMSE of 4.56 minutes. The most significant performance increase over the current model was observed under adverse weather conditions. This research advances en route delay prediction by providing more accurate delay forecasts, particularly in critical weather conditions, and proposes practical improvements to support future studies focused on enhancing model adaptability across diverse operational contexts.

With the rapidly increasing pace of urbanization and high demand for efficient modes of transport, the Urban Air Mobility (UAM) market has seen a remarkable growth in the past years. This is especially the case for the transportation of goods. Using UAM for cargo operations is likely through operating on Middle-Mile Delivery (MMD) missions to transport cargo between facilities or distribution centers in an operator’s network. The efficiency and practicality of such a network are largely affected by the selection of strategic positions for
vertiports. As vertiport location optimization is underexplored in current scientific research this paper aims to fill this research gap by developing and analyzing a multi-objective optimization model for the placement of vertiports for a middle-mile package delivery system, considering capacity, available land space, safety and noise impact factors. We develop a novel Multi-Objective Multiple Allocation Capacitated p-Hub Coverage Problem framework for an MMD UAM network and test it using the South Holland region as a case study. Notably, the model can easily be converted to other cities. First, to reduce computational efforts, the K-means
clustering algorithm is proposed. This is used to divide 6625 zones into a number of K clusters, with each
cluster representing a vertiport candidate location. Furthermore, we present a multi-objective Tabu Search
based heuristic optimization algorithm to solve the optimization problem. The impact of different factors
such as number of clusters, number of vertiports, drone range, maximum safety distance, and turn around
time The presented model provides decision-makers with the ability to assess the suitability of a region
for the implementation of a UAM MMD system and aids in the identification of potential good locations to
set up vertiports. We demonstrate that an increase in the number of vertiports leads to a higher attainable
demand coverage, however, this results in a steep drop-off in terms of safety and noise nuisance performance.
Furthermore, the results show that an increase in drone range, maximum safety distance or a decrease in turn
around time allow for overall better performing vertiport networks.

Punctuality is a key performance indicator for any airline, especially hub-and-spoke airlines, given their focus on short passenger connections. Flights that are delayed at departure need to compensate for lost time whilst airborne. Because fuelling takes place well before scheduled departure, predicted departure delays determine the planned fuel amounts for en-route speed optimization. To prevent unnecessary fuel burn, airlines benefit from highly accurate departure delay predictions. This study aims to extend previous work on airline departure delay forecasting to a dynamic and probabilistic domain, whilst incorporating novel day-of-operations airline information to further minimize prediction errors. Random Forest, CatBoost, and Deep Neural Network models are proposed for a case study on departure flights of a major hub-and-spoke airline from its hub airport between 1 January 2020 and 1 August 2023. The Random Forest model is selected for its probabilistic performance and high accuracy in predicting delays between 5 and 25 min, for which en-route speed optimization has the largest effect. At the 90 min prediction horizon, the model reaches a Mean Absolute Error of 8.46 min and a Root Mean Square Error of 11.91 min. For 76% of flights, the actual delay is within the predicted probability distribution range. Finally, this study puts a strong emphasis on explainability. Flight dispatchers are therefore provided with the main factors impacting the prediction, explaining the context of the flight. The versatility of the model is demonstrated in two shadow runs within the procedures of an international airline, where delays caused by familiar and unfamiliar factors were successfully predicted.

The European Air Traffic Management system is among the most complex systems in the world. Due to the dense nature of the European network, consequences of disruptions are often catastrophic. In particular, disruptions altering the expected flying time tend to pose great challenges to the arrival management of busy hubs. In response, EUROCONTROL released the Target Time Management (TTM) system, allowing airlines to issue Target Times of Arrival (TTA) even before depart. The TTM system helps hubs airports coordinate arrivals and departures. From the point of view of airlines, the advantage resides in being able to prioritize early arrivals of critical flights. Nevertheless, real-time prioritization is not trivial. Many studies have focused on this problem but with results limited to slot swapping in a tactical context. This is less effective compared to airlines having the ability to select a new slot at the pre-tactical level. This work covers this gap, allowing airlines to select the desired TTA even before departure. We use Deep Reinforcement Learning to create a dynamic arrival allocation model capable of prioritizing flights in terms of passenger connecting time, curfew performance, rotation delay, and fairness to other airlines. Additionally, the model is capable of adapting and react to the uncertainty in responses from the TTM. In the real-world, large anticipations in TTAs are often rejected. The model is tested with real data from SWISS International Airline. Results show an improvement of 5.9 minutes for critical passenger connection and 4.8 minutes for rotation delay versus a deterministic approach.

Assigning aircraft to gates is one of the most important daily decision problems that airport professionals face. The solution to this problem has raised a significant effort, with many researchers tackling many different variants of this problem. However, most existing studies on gate assignment contain only a static perspective without considering possible future disruptions and uncertainties. We bridge this gap by looking at gate assignments as a dynamic decision-making process. This paper presents the Real-time Gate Assignment Problem Solution (REGAPS) algorithm, an innovative method adept at resolving pre-assignment issues and dynamically optimizing gate assignments in real-time at airports through the integration of Deep Reinforcement Learning (DRL). This work represents the first time that DRL is used with real airport data and a configuration containing a large number of flights and gates. The methodology combines a tailored Markov Decision Process (MDP) formulation with the Asynchronous Advantage Actor–Critic (A3C) architecture. Multiple factors, such as flight schedules, gate availability, and passenger walking time, are considered. An empirical case study demonstrates that the REGAPS outperforms two classic deep Q-learning algorithms and a traditional Genetic Algorithm in terms of reducing passenger walking time and apron gate assignment. Finally, supplementary experiments highlight REGAPS’s adaptability under various gate assignment rules for international and domestic flights. The finding demonstrates that not only did REGAPS outperform COVID restrictions, but it can also produce considerable benefits under other policies.

Reducing uncertainty in air traffic flow management is crucial for maintaining safety and efficiency in modern aviation. In particular, forecasting Actual Take-Off Times (ATOT) for flights across Europe is challenging due to the diverse flight-specific variables and operational conditions. Additionally, to help operations, this prediction must be done well in advance in order to prevent future traffic densities from being higher than the airspace capacity. However, recent studies often make predictions on shorter horizons and do not consider the effect of knock-on delays. This study covers this gap, by focusing on larger prediction horizons and different types of delay. We enhance ATOT prediction for flights arriving at Amsterdam Schiphol Airport from European out-stations by leveraging machine learning techniques, specifically a Long Short-Term Memory (LSTM) neural network, augmented with a Multihead Attention mechanism. A model capable of capturing complex temporal dependencies and operational factors influencing the ATOT is developed utilizing data from Electronic Flight Data messages, weather reports and a EUROCONTROL dataset. The model’s performance is evaluated against traditional ensemble methods and the current Decision Support Tool (DST) system used by Luchtverkeersleiding Nederland (LVNL). Results indicate that the LSTM model outperforms existing models including a reproduction of the DST, achieving a Mean Absolute Error of 12.05 minutes at a forecast horizon of 4 hours, demonstrating significant improvements. Finally, this assessment underscores the importance of factors such as the knock-on effect in delay prediction can significantly enhance demand forecasting, leading to more efficient air traffic management and reduced delays.

The complexity of airline operations requires operations planning to be divided into multiple problems solved sequentially by the respective departments. This is particularly the case for (1) network planning and (2) maintenance planning. Despite the close interaction of these two departments, airlines typically evaluate plans from both domains separately. However, an integrated perspective is necessary to develop robust plans and effective recovery policies in this intrinsically uncertain environment. This paper presents a new modular, stochastic, discrete event simulation model of airline operations named ANEMOS (Airline Network and Maintenance Operations Simulation). ANEMOS contains both network and maintenance dynamics, allowing the evaluation of plans, policies, and scenarios from both domains. The model is validated using data from a major European airline. We show that the simulated results closely resemble the airline's historical operational performance. ANEMOS is tested with a use-case investigating the effects of adding a second reserve aircraft to a fleet of fifty wide-body aircraft. The results show that the second reserve is capable of reducing cancellations by 55%. However, such does not cover the lost revenue associated with keeping an aircraft non-operational for a part of the time.

Reactionary delays are a critical challenge in airline operations, especially within hub-spoke networks, where disruptions at spoke airports propagate and amplify throughout the fleet. Accurate prediction of these delays is essential for effective network planning, as errors can lead to flight cancellations, missed connections, and curfew infringements. However, current state-of-the-art delay prediction models do not fully integrate all elements that cause reactionary delays and affect subsequent operations. This study aims to close this gap by using a Graph Attention Network (GAT) model to predict reactionary delay distributions within a fleet network and identify the most critical flights through the analysis of attention weights. Using operational data from Swiss International Air Lines’ shorthaul fleet, the GAT model integrates node-level features, such as flight-specific parameters, and edge-level features, including rotational dependencies and passenger connections, to capture the spatial-temporal dynamics of delay propagation. The GAT model achieved reliable predictive accuracy, particularly on medium-delay days, of a root mean squared error of 15.59 minutes and a mean absolute error of 10.50 minutes. The results further reveal that the model comprehends the ripple effects caused by rotation delays. Furthermore, its attention weights confirm its capability to identify critical flights and connections, enabling the airline to allocate resources more effectively.

Aircraft maintenance scheduling is a focus point for airlines. Maintenance is essential to ensure the airworthiness of aircraft, but it comes at the cost of rendering them unavailable for operations. In current operations, aircraft maintenance scheduling must often be updated to include time for non-routine and non-schedule tasks. These non-routine tasks can increase costs, maintenance workload, and uncertainty of the airlines’ operations. This research introduces a supervised learning framework designed to forecast future non-routine task workloads accurately, improving the accuracy of the planned maintenance schedule. This framework consists of two random forest predictors which estimate the amount of non-routine tasks and the number of future work hours that should be allocated in advance for potential non-routine tasks. Our approach produces highly reliable predictions by leveraging a robust dataset obtained from an international airline. The results show an average of 20% improvement versus an existing on-site sampling method. Furthermore, our in-depth analysis of prediction distributions enables the identification of the underlying causes of significant prediction errors, shedding light on the unpredictabilities inherent to non-routine tasks.

Predicting aircraft Take-Off Weight (TOW) has been a long-standing goal for aviation stakeholders, especially for operational and regulatory bodies involved in flight planning. Accurate TOW values would enable better emissions computation, leading to more effective regulation of aviation’s climate impact. However, aircraft operators prefer to keep TOWs confidential because they are sensitive to operational trends and cost indices. Consequently, many works have attempted to circumvent this gap by predicting TOW values. Unfortunately, limited success has been achieved primarily due to the lack of accurate real-world operational data. This study is unique in utilizing operational TOW data provided by airlines. We predict TOW before takeoff based solely on Flight Plan and Terminal Aerodrome Forecast parameters, primarily focusing on flights at Amsterdam Airport Schiphol. The accuracy of several Machine Learning algorithms is directly compared. The best Mean Absolute Percentage Error of 2.17% on the Schiphol testing dataset is achieved. The model is further validated on flights at Paris- Charles de Gaulle Airport and Brussels South Charleroi Airport with errors of 4.07% and 3.41%. We found that the distribution of flights in the training dataset, particularly aircraft and airline types, significantly influenced the model’s applicability. Recommendations are also made on how to improve the model further.

Despite its success in various research domains, Reinforcement Learning (RL) faces challenges in its application to air transport operations due to the rigorous certification standards of the aviation industry. The existing regulatory framework fails to provide adequate, acceptable means of compliance for RL applications, and thus, there is no legal framework for their safe deployment yet. Guidelines must be formulated to certify RL models aimed at air transport operations to enable real-world utilisation of these promising methods. These guidelines must consider the unique characteristics of these models, deviating from the methodology of current guidelines crafted before the emergence of ML applications. The paper proposes novel certification requirements for RL models based on their technical characteristics, safety-criticality, and autonomy. This framework covers the choice of the RL algorithm and analyses the actions, agents, environment, and potential hazards and risks of the RL application. Additionally, this work outlines the evidence the certification applicant must present to demonstrate compliance with these requirements. While this framework is not a complete solution for the complex problem of certifying RL, it is intended to serve as an initial framework which can be extended in cooperation with regulatory entities.

Increasing delays and congestion reported in many aviation sectors indicate that the current centralised operational model is rapidly approaching saturation levels. Air Traffic Control (ATC) system is not expected to keep pace with the ever-increasing demand for air transportation. Its capacity is still limited by the available controllers, and the number of aircraft that each controller can manage. This system cannot be stretched any further under its current conditions. However, it is expected that the number of aircraft operating simultaneously will continue to increase. On top of this, new unmanned aviation operations promise traffic densities never seen before. The expected future increase in traffic demand has redirected focus into automated tools and alternative approaches. This research has been primarily characterised by a change in the degree of centralisation, more specifically by exploring distributed options, where control is transferred from ground-based Air Traffic Controllers (ATCOs) to each individual aircraft. As each aircraft only takes into account its neighbouring aircraft when resolving conflicts, each distributed resolution system is expected to have only a fraction of the computational strain that a centralised system would have. Nevertheless, a distributed approach has its own challenges. A crucial disadvantage is the lack of global coordination from surrounding traffic, which can affect safety. Without knowledge of the movement of intruders, decentralised solutions cannot guarantee globally optimal solutions when more than two aircraft are involved. Conflict resolution (CR) methods based on geometric solutions have proven to be very successful in achieving a high level of safety for one-to-one conflicts, where a set of rules can be defined which leads to implicitly coordinated optimal behaviour. However, at higher traffic densities, when individual conflict situations can no longer be considered isolated events, successive CR manoeuvres can lead to traffic patterns with a negative effect on the global safety. Knock-on effects of intruders avoiding each othermay result in unforeseen trajectory changes. The latter increases uncertainty regarding intruders’ future movements, decreasing the efficacy of conflict resolution manoeuvres. The goal of this research is to improve upon aircraft self-separation efficacy at higher traffic densities, with an emphasis on employing airspace designs and approaches applicable to future unmanned operations. To do so, we look at a scenario with multi aircraft interacting as a multi-agent problem. Analysis and understanding of emergent behaviour in a multi-agent environment is often almost impossible to the human eye. However, reinforcement learning (RL) techniques are often capable of identifying emerging patterns through training in the environment. We translate successful applications of RL techniques in other areas (e.g., carmobility, lane changing, freeways) to aircraft operational scenarios to mitigate the negative effect on safety...

The number of unmanned aircraft operating in the airspace is expected to grow exponentially during the next decades. This will likely lead to traffic densities that are higher than those currently observed in civil and general aviation, and might require both a different airspace structure compared to conventional aviation, as well as different conflict resolution methods. One of the main disadvantages of analytical conflict resolution methods, in high-traffic density scenarios, is that they can cause instabilities of the airspace due to a domino effect of secondary conflicts. Therefore, many studies have also investigated other methods of conflict resolution, such as Deep Reinforcement Learning, which have shown positive results, but tend to be hard to explain due to their black-box nature. This paper investigates if it is possible to explain the behaviour of a Soft Actor-Critic model, trained for resolving vertical conflicts in a layered urban airspace, by interpreting the policy through a heat map of the selected actions. It was found that the model actively changes its policy depending on the degrees of freedom and has a tendency to adopt preventive behaviour on top of conflict resolution. This behaviour can be directly linked to a decrease in secondary conflicts when compared to analytical methods and can potentially be incorporated into these methods to improve them while maintaining explainability.

Air traffic demand has increased at an unprecedented rate in the last decade (albeit interrupted by the COVID pandemic), but capacity has not increased at the same rate. Higher levels of automation and the implementation of decision-support tools for air traffic controllers could help increase capacity and catch up with demand. The air traffic control problem can be effectively modelled as a Markov game, where a team of aircraft (the agents) interact in the airspace (the environment) and cooperatively take resolution actions to achieve a common goal: safe separation in the most efficient way. As in any Markov game, the optimal policy for the team could be learnt through trial and error in a simulated environment using reinforcement learning algorithms. In this paper, we use the soft actor-critic algorithm to unravel the optimal air traffic control policy. Unlike some previous works, we propose a global (i.e., shared) reward that encourages cooperative behaviour. Furthermore, we propose a versatile policy model capable of performing heading, speed, and/or altitude resolution actions. We also demonstrate that the policy is robust and can maintain safe separation even in the presence of uncertainty regarding aircraft position, delays in implementing resolution actions, and wind. The findings of this paper also suggest that there is still significant room for improvement when controlling three degrees of freedom at the same time.

Future high traffic densities with drone operations are expected to exceed the number of aircraft that current air traffic control procedures can control simultaneously. Despite extensive research on geometric CR methods, at higher densities, their performance is hindered by the unpredictable emergent behaviour from surrounding aircraft. In response, research has shifted its attention to creating automated tools capable of generating conflict resolution (CR) actions adapted to the environment and not limited by man-made rules. Several works employing reinforcement learning (RL) methods for conflict resolution have been published recently. Although proving that they have potential, at their current development, the results of the practical implementation of these methods do not reach their expected theoretical performance. Consequently, RL applications cannot yet match the efficacy of geometric CR methods. Nevertheless, these applications can improve the set of rules that geometrical CR methods use to generate a CR manoeuvre. This work employs an RL method responsible for deciding the parameters that a geometric CR method uses to generate the CR manoeuvre for each conflict situation. The results show that this hybrid approach, combining the strengths of geometric CR and RL methods, reduces the total number of losses of minimum separation. Additionally, the large range of different optimal solutions found by the RL method shows that the rules of geometric CR method must be expanded, catering for different conflict geometries.

Current estimates show that the presence of unmanned aviation is likely to grow exponentially over the course of the next decades. Even with the more conservative estimates, these expected high traffic densities require a re-evaluation of the airspace structure to ensure safe and efficient operations. One structure that scored high on both the safety and efficiency metrics, as defined by the Metropolis project, is a layered airspace, where aircraft with an intended heading are assigned to a specific altitude layer. However, a problem arises once aircraft start to vertically traverse between these layers, leading to a large number of conflicts and intrusions. One way to potentially reduce the number of intrusions during these operations is by using conventional conflict resolution algorithms. These algorithms however have also been shown to lead to instabilities at higher traffic densities. As recent years have shown tremendous growth in the capabilities of Deep Reinforcement Learning, it is interesting to see how well these methods perform in the field of conflict resolution. This research investigates and compares the performance of multiple Soft Actor Critic models with the Modified Voltage Potential algorithm during vertical manoeuvres in a layered airspace. The final obtained performance of the trained models is comparable to that of the Modified Voltage Potential algorithm and in certain scenarios, the trained models even outperform the MVP algorithm. Overall, the results show that DRL can improve upon the current state of conflict resolution algorithms and provide new insight into the development of safe operations.

Current predictions on future drone operations estimate that traffic density orders of magnitude will be higher than any observed in manned aviation. Such densities redirect the focus towards elements that can decrease conflict rate and severity, with special emphasis on airspace structures, an element that has been overlooked within distributed environments in the past. This work delves into the impacts of different airspace structures in multiple traffic scenarios, and how appropriate structures can increase the safety of future drone operations in urban airspace. First, reinforcement learning was used to define optimal heading range distributions with a layered airspace concept. Second, transition layers were reserved to facilitate the vertical deviation between cruising layers and conflict avoidance. The effects of traffic density, non-linear routes, and vertical deviation between layers were tested in an open-source airspace simulation platform. Results show that optimal structuring catered to the current traffic scenario improves airspace usage by correctly segmenting aircraft according to their flight routes. The number of conflicts and losses of minimum separation was reduced versus using a single, uniform airspace structure for all traffic scenarios, thus enabling higher airspace capacity.

Future operations involving drones are expected to result in traffic densities that are orders of magnitude higher than any observed in manned aviation. Current geometric conflict resolution (CR) methods have proven to be very efficient at relatively moderate densities. However, at higher densities, performance is hindered by the unpredictable emergent behaviour from neighbouring aircraft. Reinforcement learning (RL) techniques are often capable of identifying emerging patterns through training in the environment. Although some work has started introducing RL to resolve conflicts and ensure separation between aircraft, it is not clear how to employ these methods with a higher number of aircraft, and whether these can compare to or even surpass the performance of current CR geometric methods. In this work, we employ an RL method for distributed conflict resolution; the method is completely responsible for guaranteeing minimum separation of all aircraft during operation. Two different action formulations are tested: (1) where the RL method controls heading, and speed variation; (2) where the RL method controls heading, speed, and altitude variation. The final safety values are directly compared to a state-of-the-art distributed CR algorithm, the Modified Voltage Potential (MVP) method. Although, overall, the RL method is not as efficient as MVP in reducing the total number of losses of minimum separation, its actions help identify favourable patterns to avoid conflicts. The RL method has a more preventive behaviour, defending in advance against nearby neighbouring aircraft not yet in conflict, and head-on conflicts while intruders are still far away.

Current predictions for future operations with drones estimate traffic densities orders of magnitude higher than any observed in manned aviation. Such densities call for further research and innovation, in particular, into conflict detection and resolution without the need for human intervention. The layered airspace concept, where aircraft are separated per vertical layer according to their heading, has been widely researched and proven to increase traffic capacity. However, aircraft traversing between layers do not benefit from this separation and alignment effect. As a result, interactions between climbing/descending and cruising aircraft can lead to a large increase in conflicts and intrusions. This paper looks into ways of reducing the impact of vertical transitions within the environment. We test two reinforcement learning methods: a decision-making module and a control execution module. The former issues a lane change command based on the planned route. The latter performs operational control to coordinate the longitude and vertical movement of the aircraft for a safe merging manoeuvre. The results show that reinforcement learning is capable of optimising an efficient driving policy for layer change manoeuvres, decreasing the number of conflicts and losses of minimum separation compared to manually defined navigation rules.